[LU-11926] Lost lease lock on migrate error

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.13.0
• Severity: 3
• Assignee: Andriy Skulysh
• Reporter: Andriy Skulysh

Description

   All the file operations have the following locking order - parent,
   child. If a lock for a child is returned to the client, the following
   operations on this file are done by the child fid.
    
   However, the migrate is an exception - it takes the lease lock first and
   takes the PW parent lock next during the MDS_REINT.
    
   At the same time, if there is a parallel racing operation (open) which
   has taken a lock on parent (conflicting with the next MDS_REINT) and
   is trying to take a lock on child - it is blocked until
   the lease cancel comes.
    
   The lease cancel is piggy-backed on the MDS_REINT RPC and is handled
   at the end of the operation, trying to take the conflicting parent lock
   first - thus a deadlock occurs.